Issues in Interpreting Associations

Question 1

Explain chance

Answer 1

• Random error.
• The possibility of observing a value or association in the sample population that is different to the true population.
• Most often reduced by increasing the sample size.

Question 2

Explain p-value

Answer 2

• Probability that a measure (value or association) from a a sample occurred by chance
• Most often used for degree of belief in the null hypothesis
• Smaller p-value = stronger evidence that observed value is real and did not occur by chance
• > 0.05 generally recognised as statistically significant

Question 3

Explain confidence interval

Answer 3

The range of values from a sample within which the “true” population value is likely to be found (95%)
Reliability of result
Close to RR =1 - less statistically significant association. Should not overlap.

Question 4

What are some the key indicators for identifying an outcome occurred by chance?

Answer 4

• Small sample size
• Large p-value
• Wide confidence intervals
• CIs include 1 or overlap

Question 5

Explain Bias

Answer 5

• Systematic error
• Source of error lies in the way the study was conducted (data collection or analysis) that distorts the association between behaviour and outcome
• Can be reduced by good study design
• Selection bias, Information bias (observer bias, responder bias, measurement bias), misclassification

Question 6

Explain Selection Bias

Answer 6

• Error due to systematic difference in the characteristics of the study group and the population from which they were selected or between comparison groups
• Sample not representative/generalisable
• Random selection and random allocation help to reduce bias

Impacted by:

• methods of selection (when certain groups are excluded)
• volunteer/self-selection
• missing data (loss to follow-up and non response)
• inclusion and exclusion criteria

Question 7

What are some key indicators for selection bias

Answer 7

• Study population clearly defined?
• Eligibility criteria
• Representative sample?
• High refusal rate or loss to follow-up?
• Cases and controls from same population?
• Selection of cases and control influenced by exposure status?

Question 8

Explain Information Bias

Answer 8

• Error due to systematic differences in the classification of exposure or outcome of study participants
• Influenced by accuracy of methods used in the study
• Observer bias: Bias introduced by the investigator. Misclassification due to knowledge of the comparison group.
• Addressed by: “blinding” - concealing exposure
• Responder bias: Bias introduced by the respondent. Re-call, reporting (creating misclassification) and non-response .
• Addressed by: blinding, shorter re-call periods or records, well worded and tested surveys

Measurement bias: Bias introduced by the tools of measurement. Unclear and inconsistent use or inaccurate measures.
- Addressed by: Testing

Question 9

What are some key indicators for information bias?

Answer 9

• Exposure and outcomes clearly defined in standard criteria?
• Data collection and entry standardised?
• Study blinded as much as possible?
• Observers/interviewers trained and supervised?
• Subjects randomised to observers/interviewers?
• Objective measurements?
• Reported information validated?

Question 10

Explain the two types of misclassification

Answer 10

Misclassification - observer and responder bias can affect the strength of association or lead to incorrect methods of association.

Non-differential misclassification: Two comparison groups (eg. exposed and unexposed) are equally likely to be misclassified. Can lead to underestimation of association but will not alter the direction of association.

Differential misclassification: Misclassification of exposure or outcome is different between the groups for comparison. Can lead to over/under estimation of associations, or false associations, can alter direction of the association.

Question 11

Explain Confounding

Answer 11

• When an independent factor distorts associations between exposure and outcome.
• Must not be on causal pathway but independent to both factors
• Confounders can provide alternative explanations for the associations observed.

Question 12

How can confounding be reduced/controlled

Answer 12

Reduced by:

• Randomization - allocation to exposure and control groups
• Restriction - limiting study to those similar to the confounder (results can’t be extrapolated)
• Matching - comparison groups have same distribution of potential confounders

Controlled (in analysis) by:

• Stratification - measuring associations between outcome and exposures separately for each stratum of confounder if known (eg. age, gender etc.)
• Statistical modelling - eg. multivariate regression analysis

Residual confounding - non-differential misclassification of a confounder, biases in same direction as confounding

Question 13

What is an effect modifier and mediator and how do they differ from a confounder?

Answer 13

Effect modifier: Factor that results in a varying association between exposure and outcome for seperate subgroups (often revealed via stratification).
Different to confounder as can be on the causal pathway and as it is a natural occurrence it must be described not controlled

Mediator: Factor on the causal pathway between exposure and outcome

Question 14

Nine Considerations for Causal Relationships

Answer 14

• Coherence: logical consistency with other information
• Analogy: similar to other cause-effect relationships
• Reversibility: reduction or removal of exposure leads to elimination or reduction of the outcome
• Strength: strong association (p-value)
• Consistency: repeatability, already observed
• Plausibility: biological mechanism for cause and effect
• Temporality: exposure occurred prior to outcome
• Specificity: relationship specific to the outcome of interest
• Dose-response: increased risk of outcome with increased exposure